Research article

Published: 2025-01-09
https://doi.org/10.20935/AcadQuant7467

Geometrical approach to logical qubit fidelities of neutral

atom Calderbank–Shor–Steane codes
Jasper J. Postema1,2,∗ , Servaas J. J. M. F. Kokkelmans1,2
Academic Editors: Misha Erementchouk, Yuhua Duan

Abstract
Encoding quantum information in a quantum error correction (QEC) code enhances protection against errors. Imperfection of quantum
devices due to decoherence effects will limit the fidelity of quantum gate operations. In particular, neutral atom quantum computers will
suffer from correlated errors because of the finite lifetime of the Rydberg states that facilitate entanglement. Predicting the impact of
such errors on the performance of topological QEC codes is important in understanding and characterizing the fidelity limitations of a
real quantum device. Mapping a QEC code to a Z2 lattice gauge theory with disorder allows us to use Monte Carlo techniques to calculate
upper bounds on error rates without resorting to an optimal decoder. In this article, we adopt this statistical mapping to predict error rate
thresholds for neutral atom architecture, assuming radiative decay to the computational basis, leakage, and atom loss as the sole error
sources. We quantify this error rate threshold pth and bounds on experimental constraints, given any set of experimental parameters.

Keywords: quantum computing, quantum error correction, Calderbank–Shor–Steane (CSS) codes, coding theory, statistical physics

Citation: Postema JJ, Kokkelmans SJJMF. Geometrical approach to logical qubit fidelities of neutral atom Calderbank–Shor–Steane
codes. Academia Quantum 2025;2. https://doi.org/10.20935/AcadQuant7467

1. Introduction
Quantum computing is currently in the Noisy Intermediate-Scale parameter that distinguishes two phases: an “ordered” phase in
Quantum (NISQ) era, in which noise imposes a limit on the which scaling up the code distance d allows qubits to be driven
fidelity of qubits and gate operations [1]. In order to build more toward arbitrarily low logical error rates, and a “disordered” phase
robust qubits, quantum error correction (QEC) has to be invoked in which errors irrevocably corrupt quantum information. This
[2]. Mapping an ensemble of physical noisy qubits to a logical mapping requires no optimal decoder to evaluate the threshold.
qubit enhances protection against errors. Surface codes are a class Entanglement is a crucial aspect of fault-tolerant computation,
of topological QEC codes that have been extensively studied [3–5]. but its consequences for error propagation are often not consid-
Only recently, it has been experimentally demonstrated for the ered in phenomological error models such as a depolarization
very first time that QEC using surface codes can suppress logical channel. Multi-qubit gates are not only pipelines for crosstalk
error rates on near-term quantum devices [6, 7]. between qubits but also inherent sources of correlated errors
A promising candidate for quantum computers is an array of neu- [15]. On a neutral atom quantum device, the instability of the
tral atoms trapped in optical tweezers. The quantum information Rydberg state that mediates entanglement, leakage outside of the
is stored in long-lived atomic hyperfine clock states [8, 9], while computational basis, and atom loss will be dominant sources of
entanglement is mediated through excitation to a high-n Rydberg errors [16]. Because they also facilitate measurement errors, their
state. This system has several attractive features, such as iden- effect on the QEC code performance is significant as the number
tical qubits, long coherence times, and a flexible geometry [10]. of cycles increases.
Recently, it has been shown that neutral atoms can be shuttled In this article, we provide a thorough analysis of the effect of en-
around using movable tweezers with excellent preservation of tanglement errors on the performance of topological QEC codes,
coherence [11], and error suppression with the Steane and toric tailored toward neutral atom quantum computers. Section 2 in-
codes has been demonstrated using this technique [12]. troduces the physics of neutral atoms and showcases error cor-
Statistical physics provides a powerful tool to analyze the per- rection protocols on neutral atom hardware. In Section 3, we
formance of error correction codes [13]. It has been shown that adopt the statistical model of errors and show how mapping is
there exists a duality between quantum codes and statistical me- achieved given a correlated error probability distribution. Results
chanics, called the statistical-mechanical mapping [14]. Mapping are given in Section 4, where a Monte Carlo simulation of noisy
quantum codes to a random Z2 lattice gauge model reveals that neutral atoms is compared to a second-order phase transition in a
the QEC error rate threshold manifests itself as a second-order statistical-mechanical model. In Section 5, we provide a summary
phase transition. In particular, the error probability is an order of our work.

1
Department of Applied Physics and Science Education, Eindhoven University of Technology, 5600 MB, Eindhoven,
Noord-Brabant, The Netherlands.
2 Eindhoven Hendrik Casimir Institute, Eindhoven University of Technology, 5600 MB, Eindhoven, Noord-Brabant, The Netherlands.
∗
email: [email protected]

ACADEMIA QUANTUM 2025, 2 1 of 14

https://www.academia.edu/journals/academia-quantum/about https://doi.org/10.20935/AcadQuant7467

2. Neutral atom error correction for arbitrary density matrices ρ and a Pauli basis {σ i }i = {X, Y, Z}.
Such a model is sufficient for many proof-of-principle simula-
Neutral atoms, such as 85 Rb and 88 Sr, often encode the qubit com-
tions, but it leaves out the intricate details of entanglement er-
putational basis states in the electronic ground-state manifold or
rors: correlated errors that may be spacetimelike separated in the
clock states. These states have long coherence times on the order
circuit. The dominant source of these errors is the erroneous im-
of seconds [8, 9], and the clock transition has a narrow linewidth
plementation of controlled logic (CZ) during the stabilization pro-
[17]. In the rest of this article, we work in the {|0i, |1i, |ri} qutrit
cess. Besides being conduits of errors, multi-qubit gates can also
manifold of the 88 Sr atom, which is endowed with the auxiliary
introduce new errors themselves, subsequently affecting other
Rydberg state |ri that facilitates long-range entanglement, which
rounds of stabilization. The order of stabilization, and whether
we leave unspecified. Figure 1 shows the level scheme of 88 Sr
all qubits are entangled simultaneously or not, will affect this
and electronic states in which the |0i- and |1i-states have been
correlation propagation and will play an important role in the
embedded. The transition from |1i to |ri is driven by a single-
transpilation of quantum circuits in the fault-tolerant era.
photon process. Though we focus on 88 Sr, our method applies to
all neutral atoms with a similar level structure. Given is a noiseless unitary evolution F (ρ) = UρU† . Introducing
a noise channel with multi-qudit Kraus representatives {Dµ }, we
The architecture of a neutral atom quantum computer provides
obtain a decoherent error channel
a scalable platform to implement QEC codes. Topological codes
X
such as the surface code, the toric code, and various color codes are E(ρ) = λµν Dµ F (ρ)D†ν (2)
favorable candidates for experimental proof-of-principle demon- µν
strations for QEC [6, 12]. Movable tweezers have been shown to
transport atoms without significant loss of fidelity [11], enabling relative to the perfect coherent channel F (ρ), for some coefficients
favorable properties such as having dedicated readout zones for [λµν ]. Through a change of basis, we can transform this channel
local measurement [18] and enabling flexible long-range entan- E(ρ) to the Pauli basis as
glement with applications to, for example, low-density parity X
E(ρ) = χµν Pµ F (ρ)Pν† , (3)
check codes [19, 20]. A great limiting factor to the fidelity of
µν
neutral atom qubits is decoherence, driven by stochastic processes
such as radiative decay and leakage, highlighted in Figure 1. The and use the Pauli twirling approximation [21–23] to obtain a
remainder of this section is dedicated to modeling these processes. channel that is diagonal in the Pµ -basis, by only keeping diagonal
terms: X
2.1. Error modeling E twirl (ρ) = χµµ Pµ F (ρ)Pµ† . (4)
µ

Often, a phenomenological depolarization model is employed to The validity of twirling is discussed in Section A.1, Supplementary
model errors, characterized by a single error parameter p ∈ [0, 1]. materials [21–23]. These Kraus operators describe decoherent
n n n n
Such an n-qubit depolarization channel L : C2 ×2 → C2 ×2 is noise channels and satisfy the completeness relation
given by
pX i i X †
L(ρ) = (1 − p)ρ + σ ρσ (1) Dµ Dµ = Im qudits (5)
3
i µ

Figure 1 • The 88 Sr level scheme, with the computational basis states encoded in the clock states (|0i = |1 S0 i and |1i = |3 P0 i). Two
important decoherence processes have been highlighted: radiative decay to the computational manifold, and leakage outside the qutrit
state submanifold. Unlabeled states depend on the choice of Rydberg state.

ACADEMIA QUANTUM 2025, 2 2 of 14

https://www.academia.edu/journals/academia-quantum/about https://doi.org/10.20935/AcadQuant7467

on the Hilbert space Hm of m qudits. The operators themselves interactions within a given plaquette to engineer a QEC code, as
are generated by single-qudit Kraus operators Eiqi acting on qubit seen in Figure 2.2
i, with index qi .
Let S = SX ∪ SZ be the group of stabilizers. Stabilization between
The first parameter our model considers is radiative decay from data qubits and stabilizer qubits S ∈ S is achieved through
the Rydberg state to |1i, with decay width γ. While the blockade local 2-qubit entanglement operations. Neutral atoms provide a
strength V is another source of infidelity, it is not of any significant multitude of implementations, such as the simple constant pulses
influence to our results since we assume V/Ω ≫ 1. For a decay proposed by Jaksch et al. [30], a time-optimal pulse proposed by
from the Rydberg state to the |1i-state, the Kraus operators read Jandura and Pupillo [31], which will be central to this Article.
    These are just a handful of examples of a very rich number of
1 0 0 0 0 0 pulse protocols developed in the past few years, such as the pulse
√
E1 = 0 1  , E 2 = 0 0 (6)
   
0 γδt proposed by Pagano et al. [32] that achieves a pulse time much
√
0 0 1 − γδt 0 0 0 comparable to that of Jandura and Pupillo [31] and has a lower
integrated Rydberg time compared to the protocol introduced by
in the qutrit basis, for small time steps δt. This decay poses a
Levine et al. [33]. Recently, also a pulse that is robust against
fundamental limit for the fidelity of multi-qubit gate operations
time-dependent variations in the laser control parameters was
on neutral atom architecture [24].
proposed by Mohan et al. [34]. Figure 3(a) provides an imple-
The second process we consider is erasure, which encapsulates mentation scheme of 5-qubit entanglement on the surface code.
both atom loss and leakage where the qubit state leaves the We implement subsequent 2-qubit gates in a clockwise fashion
computational basis through, for example, black-body radiation and assume no interaction between Rydberg-excited data qubits.
(BBR). It has been estimated that most errors in neutral atom Movable tweezers, depicted in Figure 3(b), can transport atoms
quantum computers are a result of leakage compared to Pauli within their stabilizers’ Rydberg blockade regime, where control
errors within the computational basis [25]. In the coding theory, logic can be performed, before moving them back to their initial
this is known as erasure, where the erasure symbol “?” is now ap- positions in the lattice. Figure 3(c) shows a complete schematic
pended to the binary alphabet F2 = (0, 1). It is known that (quan- overview of how QEC can be achieved on a neutral atom platform,
tum) error correction codes benefit from erasure conversion, since showing an architecture with different dedicated zones for data
erasure errors are easier to decode than X- or Z-type errors [26]. storage, entanglement, and readout.
The decay width of leakage, denoted as ω, is an implicit function
Given is a time-dependent set of 2L laser controls [z(t)] ∈ CL × RL
of the ambient temperature. We expect a tradeoff between ra-
that implements a multi-qubit entanglement operation for time
diative decay and leakage: the former imposing stricter limits on
t ∈ [0, T] on a set of L qubits arranged on a lattice Λ. Here, we
error rate thresholds than the latter. If erasure is detected using
focus on the subset of controls
ancilla qubits, as proposed in [27], we can define the erasure
channel ¯ L,


¯ ∆]
{Ωi , φi , ∆i }i∈Λ ∈ [0, Ω̄] × [0, 2π) × [−∆, (8)
I
Lerasure (ρ) = (1 − r)ρ ⊗ |0ih0| + r ⊗ |1ih1|, (7) where i denotes a qubit site, Ωeiφ are Rabi frequencies of the
2
transition between |1i and |ri, and ∆ are laser detunings of the
dependent on an erasure probability r. Thus, we implicitly assume |ri-state, which are implicitly time-dependent functions. Ω̄ and
throughout this article that we can detect and localize erasure in ∆¯ are the maximum values for the control parameters that are
real time. experimentally feasible. Let the control Hamiltonian H[z(t)] be
generated by the controls through the form
2.2. Generating entanglement
X Ωi (t) X
H[z(t)] = Xi + ∆i (t)ni + Hdrift . (9)
We briefly review QEC and the role of stabilization. Topological 2
i i
codes are a class of codes whose topological properties dictate the
error-correcting capabilities of the code. Promising examples are where Xi = |1i ihri | + h.c. addresses the single-photon transition
the surface code and the toric code. Such codes are composed between |1i i ↔ |ri i, ni = |ri ihri | is the occupancy of the Rydberg
of two qubit types: data qubits that carry the logical quantum state, and the drift Hamiltonian is given by the Rydberg-Rydberg
information in non-local degrees of freedom, and stabilizers that interactions
X C6
have to be read out to detect and recover errors. There are two Hdrift = |ri rj ihri rj |, (10)
sets of stabilizers: SX detects phase errors, and SZ detects ampli- R6ij
i>j
tude errors. CSS (Calderbank–Shor–Steane) codes are a family
summed over all Rydberg states |ri i with interatomic distance Rij
of quantum stabilizer codes [28, 29], generated by two classical
and van der Waals’ coefficient C6 .
linear codes C1 and C2 such that C2 ⊆ C1 ⊆ Fnq on the finite field
of q elements.1 This CSS construction leads to a q-ary [[n, k, d]]q - We take a Trotterization of the total unitary evolution in
code, with length n, dimension k, and minimum distance d. Each Figure 2(b), chopping controls [z(t)] into N equidistant time
underlying classical code generates a set of stabilizers that can slices of duration δt = T/N, and apply Kraus operators given
correct for either one of the error types. A favorable property of in Eq. (6) at every time step. In the limit of N → ∞, this would
surface codes is that interactions are local, i.e., the qubits can recover the exact Lindbladian evolution, though we can truncate
be laid out in a planar graph and only require nearest-neighbor for small finite N and achieve sufficient accuracy. To find the

1 F is the finite field of q elements, also called the Galois field, where q = ps is an integer 2 Note that in this article, we use a depiction of the d = 3 rotated surface code to represent
q
power of some prime number p. Fnq is equivalent to the n-dimensional field Fq × · · · × Fq . any topological CSS codes because of its simplicity.

ACADEMIA QUANTUM 2025, 2 3 of 14

https://www.academia.edu/journals/academia-quantum/about https://doi.org/10.20935/AcadQuant7467

Figure 2 • (a) The [[9, 1, 3]]2 rotated surface code, where black circles denote data qubits that carry the 2 logical degrees of freedom,
blue faces denote X-stabilizers, and golden faces denote Z-stabilizers. (b) For each stabilizer type, the relevant circuit is shown for
2-plaquettes. H is the Hadamard gate, and meas denotes a mid-circuit measurement. 4-plaquettes have analogous circuit designs.

Figure 3 • (a) Entanglement protocol for 4-plaquettes. The dashed circle indicates the Rydberg blockade radius, limited below the
interatomic distance. Controlled logic between stabilizer and data qubits is performed by transporting the neighboring qubits within
the blockade radius and applying an appropriate series of pulses before moving it back to its original position in the lattice. We
subsequently entangle other data qubits in a clockwise fashion, as indicated by the gray arrows. (b) Visualization of atom transport,
with all relevant timescales indicated below [11, 35, 36]. AOM (acousto-optical modulator) tweezers have the ability to pick up an atom
from its stationary trap and move it to a different location. Such transport allows for flexibility during execution of an algorithm, though
transpilation may pose a serious challenge [37]. (c) Schematic architecture for neutral atom quantum computing experiments, tailored
here for quantum error correction. We embed the CSS (Calderbank–Shor–Steane) code in a 2D plane of atoms (gray zone), here seen
from the side. Single-qubit operations are performed in the gray zone, entanglement is mediated in the purple Rydberg zone, and
measurements of the qubit states are performed in the green readout zone. AOMs transport atoms between layers. The Rydberg laser
that mediates the transition |1i ↔ |ri is global, as well as the laser that reads out qubit states through fluorescence. Hence, dedicated
zones are required to avoid crosstalk. Single-qubit operations in the gray zone are local and can be tuned from lattice site to site, within
realistic experimental bounds.

diagonal [χµµ ] components, we consider the Γ-matrix defined by over the modified Γ-matrix:
X
Γ= Dµ ⊗ D†µ (11)
µ Γ̄ = (I ⊗ U)Γ(U† ⊗ I), (14)
that contains all relevant information about the propagation of
errors throughout a quantum system through projections. If we where we insert U to compare to the noiseless evolution F (ρ).
expand our Kraus operators in the 5-qubit Pauli basis Because the readout time τmeas is of much greater order than all
1 X relevant decoherence timescales, we can assume that all Rydberg
Dµ = N Tr(Dµ Pν )Pν (12) states have decayed before the full measurement is concluded
2 ν
[36]. Later on, we will see that this gravely affects the logical error
to obtain the E twirl (ρ) channel (4), the diagonal [χµν ] components rate bounds. More details of the precise step-by-step calculation
follow consequently from the relation are elaborated on in Section A.2, Supplementary materials.
 
1
χµν = N Tr Γ̄ · (Pµ ⊗ Pν ) , (13)
4

ACADEMIA QUANTUM 2025, 2 4 of 14

https://www.academia.edu/journals/academia-quantum/about https://doi.org/10.20935/AcadQuant7467

3. Statistical model of errors is the quotient group of e over the stabilizer group S. Trivial errors
are in the same equivalence class as the 0-error. The statistical
The statistical-mechanical mapping is a mathematical duality
mapping must abide by this same gauge symmetry.
between the performance of QEC codes and a specific family of
random bond Ising models or random plaquette gauge models. Our focus is to find a classical Z2 gauge theory model whose
This duality was first realized by Dennis et al. [14] and later thermodynamical properties mimic the behavior of errors on
generalized and applied for correlated noise [38–40]. Because of topological codes. We assign a classical spin degree of freedom
the statistical nature of errors, it is very natural to borrow tools si ∈ Z2 = {↑, ↓} to each stabilizer qubit site, and map data
from statistical physics to analyze the behavior of error propaga- qubits to bonds or plaquettes whose strength (±J) is dictated by
tion within certain noise models, and their impact on the overall the underlying error configuration, as depicted in Figure 4. The
performance of the code. This gives us a concrete way of predicting gauge spin model must satisfy a few properties:
upper bounds on error rate thresholds without using any decoder,
such as the Blossom algorithm [41] or belief propagation [42]. 1. Topology compatibility—It must be consistent with the un-
derlying topology and boundary conditions of the quantum
First, we address the gauge symmetry of QEC briefly. An error code. For instance, the dual Hamiltonian of the toric code
configuration e is said to be homologically trivial if there exist must match its topology: T 2 = S 1 × S 1 for the 2D toric code
Q
a subset of stabilizers S⋆ ⊆ S such that e = s∈S⋆ s. An error and T 3 = S 1 × S 1 × S 1 for the 3D toric code.
configuration leads to logical failure if and only if the recovery
r leads to an operator e ⊕ r that does not commute with every 2. Configuration averaging—The distribution of the random
single of the 2k logical operators of an [[n, k, d]]q -code. In other interactions must mimic error correlations as predicted by
words, a decoder fails if and only if it fails to identify the right the twirled Pauli channel. The error probability p determines
homology class of the underlying error e. Let ē be the equivalence the probability of a local anti-ferromagnetic coupling.
class of errors that are in the same homology class as some error 3. Gauge symmetry—Errors that are equivalent to each other
configuration e, i.e., within the homology class Eq. (15) must map to the exactly
same family of Hamiltonians.
ē = {e + S | S ∈ S} (15)

Figure 4 • (a) A CSS (Calderbank–Shor–Steane) code is mapped to a spin gauge model with random bonds, with spins sX (blue)
and sZ (golden) representing stabilizers and bonds representing data qubits. Bonds between spins si are ferromagnetic (+J) and favor
long-range order, if the corresponding data qubit does not have an error that can be picked up by the stabilizer type i. Such bonds are
depicted as blue dashed lines. (b) If a data qubit undergoes an error (red), triggering two flags (pink), its respective bond turns locally
anti-ferromagnetic (−J), indicated by a red dashed line.

ACADEMIA QUANTUM 2025, 2 5 of 14

https://www.academia.edu/journals/academia-quantum/about https://doi.org/10.20935/AcadQuant7467

We aim to find a dual statistical-mechanical Hamiltonian H(⃗si ). model has the gauge symmetry of the stabilizers built in, which
Let the bond strength be defined as ηJ, with magnitude J and is invariance under multiplication with elements of the stabilizer
probabilistic sign η.3 Let π(ē) denote the probability that a certain group:
error configuration belongs to the homology class of e. Then, the e 7→ eS, (20)
Nishimori conditions [43] are the set of constraints that yield the
for S ∈ S, and generalizes for any underlying error model. For
right Hamiltonian parameterization such that
CSS codes with independent X and Z errors, we can separate this
Ze [η, J] = π(ē), (16) Hamiltonian into two independent ones, pertaining to each error
type. This gravely simplifies our analysis.
where Ze is the partition function of the gauge Hamiltonian. The We can generalize the Hamiltonian formalism for arbitrary di-
statistics match exactly since the Hamiltonian is invariant under mensions. The D-dimensional isotropic partition function Z[J, η]
the gauge symmetry generated by the stabilizers S ∈ S. dropping the label e, corresponding to a toric code with one single
At low temperature T, the system favors global order. Increasing T error type occurring at uniform rate p, is given by
allows more thermal fluctuations to disorder the system, up until a  
critical temperature Tc above which all global order is broken. This X X
Z[J, η] = exp βJ ηij si sj , (21)
second-order phase transition marks a phase boundary between
{⃗s} ⟨i,j⟩
order and disorder. The intersection point with the Nishimori
condition yields the threshold error rate pth below which QEC D
where {⃗s} = {−1, 1}×d is the set of all possible spin config-
displays order, i.e., can drive the logical error rate arbitrarily P
urations, ⟨i,j⟩ is the sum over the nearest-neighbor sites only,
close to 0 if the code distance d is scaled up sufficiently high. β = T is the inverse temperature, ηij ∈ {−1, 1} is the coupling
−1
The Nishimori point, where the Nishimori line crosses the phase between neighboring spins i and j, and J is the interaction strength
boundary, is a renormalization group fixed point [44]. set to unity. It is easy to see that the Z2 gauge symmetry is given
From now on, we make a distinction between three error types. by
Pauli errors occur at a rate p and signify errors within the compu- si 7→ σi si , ηij 7→ σi σj ηij , (22)
tational basis. Three distinct types are considered: X (amplitude) for some set of gauge variables σi ∈ Z2 . In this model, we
and Z (phase) errors are chosen to be elementary errors; a Y error can define an order parameter m2 that distinguishes the ferro-
is simply the occurrence of both types of errors at a single site: magnetic and paramagnetic phases, given by the mean-squared
Y = ZX. 4 M errors are measurement errors, appearing at a rate magnetization
q. We make this explicit distinction between spacelike errors and m2 = lim hsi sj i. (23)
timelike errors to paint contrast between (2+1)D and 3D QEC |i−j|→∞

codes. Finally, we have erasure errors denoted by the erasure For T < Tc , m > 0, and for T > Tc , m2 = 0. The critical tem-
2

symbol “?”, at a rate r. perature Tc that marks the phase boundary can be derived from
a multitude of quantities, such as a divergence in the magnetic
3.1. Z2 Gauge theory—Random bond Ising model susceptibility.

Here, we construct a dual Hamiltonian on a d × d lattice of For an independent uniform error model, the Nishimori condition
data qubits, where d is also implicitly the measure of the code is given by  
distance of the topological code in which we encode our quantum 1 1−p
J= ln (24)
information. Let P 1 denote the set of single-qubit Pauli operators 2β p
for a depolarization probability p. Along the Nishimori line, the
P 1 = {I, X, Y, Z}, (17) free energy of the ferromagnetic and anti-ferromagnetic regions
are identical, yielding enhanced symmetry. For D = 2 and D = 3,
where I denotes the single-qubit identity operator (unless speci-
Monte Carlo simulations [43, 45] predict a second-order phase
fied otherwise by a subscript index). Then, P N denotes the N-qubit
NN transition for errors at constant rate p, at threshold error rates of
set of Pauli strings: P N = n=1 P . Let the scalar commutator
1
N
×2N
[[·, ·]] : P N × P N → C over two operators O1 , O2 ∈ C2 be p2D and p3D (25)
th ≈ 0.109 th ≈ 0.250,
defined as follows:
1 h i which are in agreement with classical simulations of the toric code,
[[O1 , O2 ]] = N Tr O1 O2 O†1 O†2 . (18) such as those provided by the stim Python package [46]. The sur-
2
face code has a lower threshold because the boundary conditions
Then, in its most general form, the dual Hamiltonian reads make it less likely for the decoder to correctly identify the right
X Y equivalence class of certain error configurations, and faithfully
He [⃗s] = − Jj (σ)[[σ, e]] [[σ, Sk ]]sk , (19)
correct for them. Toric code simulations therefore provide an
j,σ∈Pi k
upper bound on the performance of the surface code.
where J determines the coupling strength, [[σ, e]] is the signum
Q
function that determines the sign of the coupling, and k · · · is 3.2. Z2 Gauge theory—random plaquette model
the product of spins connected by the coupling [38]. This Ising
QEC experiments usually involve measuring syndromes over a
multitude of cycles. Since measurements are not guaranteed to
3 The signum function η is a function that tells us the sign of a quantity, such as a +1 coupling
strength for no error, 0 for an erasure, and −1 for a Pauli error.
4 Note that we dropped the global phase i here, which yields Y† = −Y.

ACADEMIA QUANTUM 2025, 2 6 of 14

https://www.academia.edu/journals/academia-quantum/about https://doi.org/10.20935/AcadQuant7467

produce the desired outcomes, either due to uncertainty in the that is, neither bond nor site percolation is violated globally. In
measurement procedure itself or due to an amplitude error oc- Figure 5, an example configuration of percolation is displayed.
curring on a stabilizer qubit after stabilization, a number of cycles Calculating the resulting error rate threshold rth is well known
on the order of d are run to make sure that an error history can in literature and depends solely on the underlying geometry of
be faithfully traced back. Random plaquette gauge models map the lattice [52, 53]. The tiling of a qubit lattice that embeds a
directly to such (2+1)-dimensional spatiotemporal QEC codes [39, topological quantum code will therefore affect this threshold.
47]. The corresponding Hamiltonian is a generalization of Eq. (19) The order parameter associated to percolation is the percolation
to 2+1 dimensions, such that strength: the fraction of sites Π that are a part of an infinite cluster:
X Y (
He [⃗s] = −J τP si , (26) 0 if r < rth ,
Π= (32)
P P∋i 1 if r > rth .
where P are plaquettes with signs {±1} and P ∋ i are edges that Both left-right and top-down percolations are required for in-
are adjacent to the plaquette P. The gauge transformation is formation to be protected on the toric code. In Figure 6, the
temporal effects of erasure are depicted.
si 7→ σi si , τP 7→ σi σj σk σl τP , (27)

for some set of gauge variables σi ∈ Z2 , and {i, j, k, l} ∈ P. The 3.4. Capturing correlations
order parameter is given by the Wilson loop
In order to insert the results from the Pauli twirling calculations,
Y we can randomly draw local 5-qubit Pauli error configurations
W[τ ] = sl (28)
l∈τ

over a closed loop τ over the lattice, where sl ∈ Z2 sit on

edges. This quantity is gauge-invariant and non-local, providing
an excellent candidate to quantify phase transitions, according to
Elitzur’s theorem [48]. Let |τ | be the length of the loop and Sτ the
enclosed area. In the ordered phase, the thermodynamical average
of the Wilson loop is given by

hW[τ ]i ∝ exp (−c1 |τ |) , (29)

while in the disordered phase, we find

hW[τ ]i ∝ exp (−c2 Sτ ) , (30)

where c1 and c2 are smooth functions of the temperature [49,

50]. These scaling laws are called the perimeter law and area
law, respectively. We stress that a 3D random bond Ising model
does not describe measurement errors (because we are implicitly
solving ∂ ⋆ u = 0 for error chain 1-forms u, where ∂ is a total
differential and ⋆ is the Hodge star operator, indicating that we
must map errors on data qubits to plaquettes). However, if there
is no z-coupling, i.e., no measurement errors, it was proven by
Suzuki [51] that this model reduced to individual copies of a 2D
random bond Ising model. In particular, a threshold error rate of
(2+1)D
pth ≈ 0.033 (31)

was established for a uniform isotropic error model [39], at the

intersection with the same Nishimori line generated by Eq. (24).

3.3. Erasure model—percolation

Figure 5 • (a) Example of percolation on a 15×15 spin system
Erasure of data qubits is equivalent to the problem of bond per- that is dual to the toric code, where spins are given by gray
colation, while erasure of stabilizer qubits is equivalent to site nodes. Bonds and vertices are randomly removed at a uniform
percolation. The erasure of stabilizers poses the most stringent rate r = 0.15, leaving a subgraph on which top-bottom and left-
danger to the existence of a logical qubit, since the erasure of a right percolation can be found. Examples of logical support are
stabilizer automatically erases all of its bonds. Additionally, they indicated in green and blue, respectively. As periodic boundary
are the only classical feedback we can infer from the system, conditions apply, percolation clusters must connect left-right and
top-bottom on the same horizontal and vertical lines, respectively.
so losing them gives us access to only part of the error history,
(b) Percolation probability as a function of the fraction of unerased
affecting the threshold error rate.
qubits, for system sizes L = 5, 10, 25. Darker colors indicate larger
Erasures do not corrupt quantum information if and only if all log- lattices. The vertical black line indicates the threshold at around
ical degrees of freedom are shielded from the effects of erasures, 0.75 (r ≈ 0.25).

ACADEMIA QUANTUM 2025, 2 7 of 14

https://www.academia.edu/journals/academia-quantum/about https://doi.org/10.20935/AcadQuant7467

Figure 6 • Example of the presence of erasure errors erasing data qubits (bonds) and stabilizer qubits (vertices) from the lattice. Time
moves from left to right. Each panel shows the lattice at a later cycle at regular temporal intervals. Erased data qubits will remain
erased throughout the entire duration of the quantum error correction experiment, though stabilizers may be refreshed depending on
the availability of fresh atoms.

according to its underlying distribution function and apply them from that readout. Because neutral atom quantum computers typ-
to all unit cells. Motivated by this structure, we partition the qubit ically suffer from long measurement timescales, the measurement
lattice Λ into a non-disjoint set of sub-lattices {Λi }i such that time τmeas dictates the clock speed of error correction. Since deco-
[ herence times are typically of lower orders than τmeas , we assume
Λ= Λi (33) all Rydberg states have emptied at the end of every cycle, which
i
drastically impacts the influence of the CNOT (controlled-NOT
that are necessarily non-intersecting, i.e., Λi ∩ Λj 6= ∅ does not gate) entanglement protocol on the error rate threshold.
imply i = j, which is a general requirement regardless of the The rest of this section is dedicated to obtaining qubit fidelities for
code geometry and connectivity. If all sub-lattices were disjoint, the toric code implemented on a neutral atom quantum device,
then the model would resemble an independent error model after assuming two major loss channels mediated by the fragile Ryd-
sufficient coarse graining. For every stabilizer Sµ ∈ S, we define berg state. We evaluate the local error channels for 4-plaquettes,
the local neighborhood as all data qubits that interact with the extracting probabilities χµµ for certain twirled Pauli error strings
stabilizer, i.e., on the data qubits, which includes the probability q of a false flag
ΛSµ = {ν | Hµν = 1}, (34) error on the stabilizers. We insert these into a 3D random pla-
where Hµν is the parity check matrix of the underlying CSS code. quette gauge model. Additionally, erasure rates are incorporated,
The Nishimori conditions for a general correlated error model are whose effects are a result of the integrated Rydberg lifetime of
given by the qubits plus the finite lifetime of the tweezer traps. Because
1 X
of the anisotropy of our system, spacelike and timelike Wilson
βJi (σ) = ln ϕi (τ )[[σ, τ −1 ]], (35)
|PΛi | τ ∈P loops have different sensitivities to variations in temperature.
Λi
Our calculations have shown that disorder is more spacelike than
where the functions Ji : PΛi → R are now defined for local timelike, so that we sample timelike Wilson loops only. The details
neighborhoods [38], which are the 5-qubit 4-plaquettes of the of the Metropolis algorithm and handling phase transitions are
toric code, but can be generalized for any topological CSS code presented in Section 3, Supplementary materials [54, 55].
such as color codes.
4.1. Twirled probabilities
4. Code performance analysis Under the Pauli twirling approximation, we can find probabilities
Now we analyze the performance of the toric code over a range associated to a certain Pauli error Pµ occurring on a local cluster
of decay rates γ and ω, and quantify how the QEC error rate of 5 qubits, by calculating the diagonal terms χµµ , for different
threshold depends on their interplay. For several discrete points pulse implementations. The two implementations we consider in
in the (γ, ω)-parameter space, we will calculate the error rate this article are the Jaksch protocol [30] (also colloquially referred
threshold pth . The twirled channel E twirl (ρ), as given in Eq. (4), to as the π − 2π − π pulse) and the time-optimal Jandura
gives us the likelihood of certain local error configurations as well protocol that minimizes the Rydberg population time [31]. We
as the likelihood of a measurement error with probability q. The inherently assume in our models that non-leakage decay from
random plaquette gauge model described in Section 3.2 allows us the Rydberg state goes to the |1i-state, so that errors during the
to evaluate the performance of QEC codes over different cycles c CZ-implementation can only corrupt phase information. Thus, we
by applying the twirled quantum channel: find that X-stabilization yields X-errors only, and likewise for Z-
stabilization. This way, we split our analysis up into a competition
z
c times
}| { between independent X/Z Pauli errors versus erasure errors.
E (c) (ρ) = E twirl ◦ · · · ◦ E twirl (ρ). (36) Because of the gauge symmetry of topological quantum error-
correcting codes under the group of stabilizers S, we simplify the
To effectively combat measurement errors, we perform d rounds
analysis by grouping probabilities in terms of error count. For data
of measurements for a d × d toric code. Additionally, it is common
qubit errors on general n-plaquettes, this corresponds to grouping
practice to assume that the final round of measurements is perfect,
by reading out every data qubit and inferring a parity syndrome P(k errors) and P(n − k errors), (37)

ACADEMIA QUANTUM 2025, 2 8 of 14

https://www.academia.edu/journals/academia-quantum/about https://doi.org/10.20935/AcadQuant7467

since they are congruent to each other under applying the stabi- 4.2. Handling erasure
lizer operator associated to the ancilla qubit that stabilizes them.
For the erasure rate ω, we consider two loss processes. Leakage,
The results are displayed for 4-plaquettes in Figure 7. A crucial
as described in Section 2, is a result of BBR-mediated transitions.
feature of these simulations is that the measurement error rate
Qubits can be erased with a probability proportional to the time
is usually on the same order of magnitude as non-measurement
spent in the Rydberg state and the black-body scattering width.
error rates. This means that in the limit of very strong decoherence
The standard value we adopt is ΓBBR = 2π · 840 Hz. Another loss
γΩ ≈ 1, the underlying spin model starts to resemble more a 3D
process is a consequence of the finite lifetime of the trap, which is
isotropic random plaquette gauge model than a 2D random bond
the dominant mechanism behind erasure. Traps are described by
Ising model, with a suppressed error rate threshold (Figure 8).
a lifetime Ttrap , with an ensemble of atoms knowing exponential
Importantly, from Figure 7, we retrieve fidelity behavior that is decay. Assuming a measurement time on the order of τmeas ≈ 1 −
consistent with heuristic scaling laws. For the Jaksch protocol, the 10 ms, and a trap lifetime on the order of Ttrap ≈ 10−50 s, we have
curves follow the a trap ejection rate that increases monotonically as a function of
the number of cycles c.
lim p = aγ F (38)
γ→0
One major challenge of the statistical mapping is calculating
scaling law at low γ-values for some a ∈ R and F ∈ {1, 2}
+ Wilson loop averages on a 3D plaquette model with percolation.
denoting how many data errors occurred. This is consistent with The loop τ cannot contain any erased qubits as that would auto-
the intuition that 1 decay process should occur with probability matically set its expectation value to 0, and the enclosed area Sτ is
proportional to γ, while 2 decay events occur at a rate propor- calculated by taking into account the geometry of the holes that are
tional to γ 2 . For the time-optimal protocol, the tails of one- and encircled by the loop. For higher erasure rates, the shape of Wilson
two-error probabilities both scale with γ. The intuitive picture loops becomes more convoluted, marking a major roadblock in
behind this scaling is that since we drive collective multi-qubit our computations for high r. For this reason, we extrapolate to
excitations to the Rydberg state, double errors occur at a rate γ → 0, knowing that for the final timeslice, percolation on a
1 − (1 − γδt)2 ∼ O(γ) for small timescales δt. random bond Ising model in 2D gives thresholds at r = 0.5 and
r ≈ 0.25 for bond and site-bond percolation, respectively, the
latter of which is established in Figure 5. At the final timeslice, we

Figure 7 • The error probability p of a certain number of errors that can occur on 4-plaquette data qubits as a function of the decay
width γ in units of the Rydberg laser Rabi frequency Ω, given by a sum of elements χµµ from Eq. (3). Here, no leakage or erasure is
assumed. M in the legend indicates that a measurement error occurred (on the stabilizer qubit). Simulations were performed by using
the finite element Kraus evolution for both pulse protocols, with N = 20 Trotterization steps for every single 2-qubit entanglement
pulse. In the small error limit γ/Ω ≪ 1, we retrieve scaling laws compatible with heuristic arguments as highlighted in the main text.
For 2-plaquettes, we retrieve similar plots and scaling laws, omitted for clarity’s sake. For either protocol, the curves for “2 errors” and
“2 errors + measurement error” are overlapping. For the time-optimal protocol, the curves for “1 error” and “no errors + measurement
error” overlap. The latter protocol is also completely symmetric with regard to X ↔ Z.

Figure 8 • Elaborate example of different 2-qubit correlations on a 4-plaquette, for the time-optimal protocol at γ/Ω = 0.1. The order
of stabilization starts from the top left and rotates clockwise. (Light)blue bars indicate the correlations between erroneous red qubits on
the (light)blue plaquettes. Note that the first blue bar is missing because we only look at causal correlations between the second qubit
being erroneous, and any qubit later in the stabilization process being erroneous as well.

ACADEMIA QUANTUM 2025, 2 9 of 14

https://www.academia.edu/journals/academia-quantum/about https://doi.org/10.20935/AcadQuant7467

must have logical support for both logical qubits. We can imagine total erasure rate is given by
two scenarios: 1
ω = fint ΓBBR + , (39)
Ttrap
• Lost qubits are irrevocably lost.
where fint is the fraction of the integrated Rydberg time compared
• Lost stabilizer qubits are refilled in the next cycle with atoms
to the time it takes to conduct 1 cycle. We adopt this conven-
pulled from a fresh reservoir, which is maintained and mon-
tion for later purposes. For infinitely stable traps (Ttrap → ∞),
itored throughout the entirety of the experiment to ensure
the maximum number of cycles c⋆ is bounded by BBR-induced
a near 100% filling rate. Data qubits will never be refilled
scattering during excitations to the Rydberg state, and will scale
since erasures pose less stringent bounds on QEC than Pauli
like c⋆ ∼ 1/ΓBBR . In this theoretical case, we expect a lifetime
errors.
upper bound of ≈ 3.5 · 103 cycles. From trap lifetimes of the
order of minutes, we already see that the lifetime of a logical qubit
The latter will, of course, benefit QEC error rate thresholds, at outlasts the physical qubit lifetimes. Therefore, we can establish
the cost of more qubit resources and continuously monitoring a that erasure plays only a minor role in the lifetime of logical qubits
reservoir in parallel. as long as traps are sufficiently long-lived.
Sweeping over (γ, r̄), we retrieve a set of phase diagrams, dis-
4.3. Logical qubit fidelity and lifetime
played in Figure 9, for the two pulse implementations consid-
First, we provide rough bounds on the lifetimes of logical qubits ered. The diagrams represent the intersections of the Nishimori
as predicted by our model. We adopt a set of parameters as sum- sheets and the p − T diagrams generated by sweeping over a
marized in Table 1. The most dominant factor that impedes arbi- range of p and locating its associated critical temperature Tc . The
trarily long quantum memory coherence times is erasure. Because Nishimori sheet is calculated using Eq. (19) and equates to
lost data qubits cannot be replaced, the percolation threshold |P|
marks the absolute end of the logical qubit. Taking both atom loss T(p, r) = P , (40)
τPauli ̸=0 [[σ, τ
−1 ]]ln(ϕ
j (τ )(1 − r))
and leakage into account, with a probability proportional to the
integrated Rydberg time, we estimate the number of cycles that under the identification that an erasure error is given by τ −1 = 0.
an experiment can maintain before losing the logical qubit. The The effect of the erasure error rate r drops out of the Nishimori

Figure 9 • (a) The phase diagrams for our quantum error correction implementation of the toric code, with stabilizer circuits being
implemented by either the Jaksch or time-optimal protocol. The dark blue regions indicate the QEC regime where arbitrarily low logical
qubit fidelities can be achieved, granted d is sufficiently large. The light blue regions indicate regions where QEC is only viable if lost
stabilizers are replenished from some fresh reservoir, alleviating some of the fidelity bounds. The blank regime entails all parameters
(γ, r̄) such that QEC is no longer possible. Error bars are a combined results of both standard Monte Carlo finite sampling errors for
2500 error distributions and 25,000 steps, and the uncertainty in γ due to randomly distributing errors over a partially erased lattice
yielding a different effective error rate standard deviation. Because Wilson loop calculations for r ⪆ 0.35 are very difficult to perform
in a Monte Carlo setting, we use a fit to approximate the phase boundaries, using the knowledge that for the limiting case of erasure-
only errors (γ → 0), we must recover 2D percolation thresholds calculated in Section 3.3. A set of realistic system parameters are
highlighted in black dashed lines/gray regions as examples. (b) Typical graph of the behavior of Wilson loop averages, normalized by
perimeter/area. Each line is the average of 25 randomly chosen runs averaged over 10,000 steps each, at r = 0 and γ/Ω = 10−3 , for
both the ordered (T < Tc ) and disordered (T > Tc ) regimes, displaying a clear transition from a perimeter to an area law. The left graph
shows − ln W[τ ]/|τ | on the y-axis, while the right shows − ln W[τ ]/Sτ . (c) Examples of when Wilson loops can be easily constructed.
If the loop encloses no erased qubits, the enclosed area Sτ is equal to a Euclidean area. If some erasures are enclosed, then the area is
effectively less. For larger r̄, the calculation of the area/perimeter law for arbitrary loops becomes convoluted and imprecise.

ACADEMIA QUANTUM 2025, 2 10 of 14

https://www.academia.edu/journals/academia-quantum/about https://doi.org/10.20935/AcadQuant7467

Table 1 • Realistic experimental parameters. These values are Using the phase diagrams, one can check for any given set of
not fixed, but they serve as a practical reference throughout the experimental parameters whether QEC can be achieved, and what
remainder of this article the lifetime of the qubit is with respect to the maximum number
Parameter Value Units Description of measurement cycles c (which is equal to the maximum code
Ω 5–30 MHz − distance d). Of course, the model can be extended to study any
γ 96.5 µs Lifetime of the 61s state [34]. distance-d topological CSS codes and even to codes where the
ΓBBR 5.3 kHz All BBR transitions for number of cycles is different from the code distance, to allow for a
n ∈ [45, 75] [56]. more precise characterization. More generally, our model allows
Ttrap 10–50 s − for any pulse protocol to be studied under any noise type.
τmeas 1–10 ms −

5. Summary
condition through cancelation of logarithms, which makes sense
Physical implementations of QEC on real hardware facilitate the
physically considering it defines the distribution of wrong sign
propagation of correlated errors throughout the lattice of qubits
plaquettes on the plaquette gauge model, which is only well de-
in which the QEC code is embedded. This effectively lowers the
fined on plaquettes that haven’t yet been erased.
threshold error rate pth , below which increasing the code distance
In Figure 9, we have rescaled the r-axis and denote with r̄ the d will suppress the logical error rate. In this article, we have
effective erasure probability, defined by the relation adopted the statistical mapping to characterize the entanglement
strength of the propagation of errors for a neutral atom quantum
r̄ = 1 − exp (−ωτmeas c) , (41) computer, based on the level scheme of 88 Sr, adopting physical
implementations of laser pulses that realize multi-qubit gates.
so that our phase diagrams are agnostic to measurement duration
We unified this statistical model with the percolation model of
and the precise number of cycles/code distance d. The interpreta-
erasures to obtain one overarching model that can handle all error
tion of this parameter is the average fraction of lost qubits at the
types simultaneously. Under this mapping, we located second-
final timeslice of the experiment, which is what we must compare
order phase transitions that correspond to QEC error rate thresh-
to rth calculated in Section 3.3. By adopting this parameter, we
olds, quantifying the interplay between erasure and Pauli errors,
can change the erasure probability and the number of cycles
which can be used to gauge the effectiveness of QEC for a given set
without having to recalculate the diagram for these new specific
of experimental parameters.
parameters.
We observed that Rydberg-related errors on a neutral atom sys-
Figure 9 reveals the consequences of the underlying neutral atom
tem, under the assumption that the measurement time τmeas is
physics to large-scale QEC. One aspect that is clearly captured
of larger magnitude than typical decoherence times τdec , typically
in these results is that erasure conversion is indeed a favorable
introduce more spacelike errors than timelike errors, so that the
protocol for QEC, as evident from the asymmetry of the diagrams.
corresponding statistical model resembles more a 3D plaquette
For more information, we refer the reader to [25, 57]. For 88 Sr,
gauge model than a 2D depolarization model with long-range
we have highlighted some experimental parameters for trap life-
spatial error correlations. We have also shown than time-optimal
times in dashed lines, showing that strontium-based neutral atom
pulses give slightly more leeway for QEC than simple pulses do
quantum computers are already suitable for QEC experiments
such as a π − 2π − π pulse, proving indeed they are more suited
pertaining to quantum memory on toric codes. We find thresholds
for QEC implementations, though the difference may only be
in the no-erasure limit of
sufficiently relevant for NISQ-era experiments.
Jaksch TO
γth ≈ 2.5 · 10−3 Ω and γth ≈ 4.5 · 10−3 Ω, (42) In our model, similar to [27], we assume that erasure can be
tracked in real time through the use of ancilla qubits. This intro-
which are consistent with literature on optimizing neutral atom duces a qubit overhead on the order of the number of physical
laser protocols for stabilization for QEC [23], as it provides an qubits n. We cannot ignore the fact that more efficient overhead
upper bound on error rate thresholds without requiring an (opti- would be possible in the future. Moreover, we have only studied
mal) decoder. Note that the results agree very well because of the the error rate thresholds in this article. Future work could also
simplicity of the toric code, having local connectivity, and is com- focus on the exponential suppression of the logical error rate.
patible with the range of state-of-the-art experimental parameters
Further research could investigate more extensive error models,
10−4 Ω ⪅ γ ⪅ 10−3 Ω. (43) including sources like stray fields or classical laser noise, the latter
of which was studied in [58]. Furthermore, one can investigate
the effect of simultaneous 5-qubit stabilizer pulses on code per-
The errors in γ from our simulations are of a higher order of
formance. For more discussion on this, we refer the reader to
magnitude than those produced by simulations such as provided
[23]. Lastly, our analysis can be generalized to systems where data
by the stim Python package, though we stress that the rough order
qubits interact through their mutual Rydberg–Rydberg interac-
or magnitude of an error rate threshold is more important than
tions, so that spatial correlations are more prone to occur.
a −5% variance. For more complex topological codes with more
intricate error models, however, these minor Monte Carlo errors
may be acceptable trade-offs, as our model may achieve greater
time efficiency compared to performing many rounds of decoding
Acknowledgments
which can be slow and underestimate the maximum error rate We thank Leo Radzihovsky, Jasper van de Kraats, Robert de
threshold. Keijzer, Raul Parcelas Resina dos Santos, Denise Ahmed-Braun,

ACADEMIA QUANTUM 2025, 2 11 of 14

https://www.academia.edu/journals/academia-quantum/about https://doi.org/10.20935/AcadQuant7467

and Emre Akaturk for fruitful discussions. This research is fi- Publisher's note
nancially supported by the Dutch Ministry of Economic Affairs
Academia.edu Journals stays neutral with regard to jurisdictional
and Climate Policy (EZK), as part of the Quantum Delta NL pro-
claims in published maps and institutional affiliations. All claims
gramme, and by the Netherlands Organisation for Scientific Re-
expressed in this article are solely those of the authors and do
search (NWO) under Grant No. 680.92.18.05. This project is par-
not necessarily represent those of their affiliated organizations, or
tially funded by the European Union through the Horizon Europe
those of the publisher, the editors, and the reviewers. Any product
programme HORIZON-CL4-2021-digital-emerging-01-03 via the
that may be evaluated in this article, or claim that may be made by
project 10170144 (EuRyQa).
its manufacturer, is not guaranteed or endorsed by the publisher.

Funding Copyright
The authors declare no financial support for the research, author-
©2025 copyright by the authors. This article is an open ac-
ship, or publication of this article.
cess article distributed under the terms and conditions of the
Creative Commons Attribution (CC BY) license (https://creative
commons.org/licenses/by/4.0/).
Author contributions
Conceptualization, J.J.P.; methodology, J.J.P.; software, J.J.P.;
validation, S.J.J.M.F.K.; formal analysis, J.J.P.; investigation, References
J.J.P.; data curation, J.J.P.; writing—original draft, J.J.P.; 1. Preskill J. Quantum computing in the nisq era and beyond.
writing—review and editing, J.J.P. and S.J.J.M.F.K.; visualiza- Quantum. 2018;2:79. doi: 10.22331/q-2018-08-06-79
tion, J.J.P.; supervision, S.J.J.M.F.K.; project administration,
S.J.J.M.F.K.; funding acquisition, S.J.J.M.F.K. Both authors have 2. Terhal BM. Quantum error correction for quantum memo-
read and agreed to the published version of the manuscript. ries. Rev Mod Phys. 2015;87(2):307. doi: 10.1103/RevMod
Phys.87.307

3. Kitaev A. Fault-tolerant quantum computation by anyons.

Conflict of interest Ann Phys. 2003;303(1):2. doi: 10.1016/S0003-4916%2802
The authors declare no conflict of interest. %2900018-0

4. Fowler AG, Mariantoni M, Martinis JM, Cleland AN. Sur-

Data availability statement face codes: towards practical large-scale quantum compu-
tation. Phys Rev A. 2012;86(3):032324. doi: 10.1103/Phys-
Data supporting these findings are available within the article, at RevA.86.032324
https://doi.org/10.20935/AcadQuant7467, or upon request. Sim-
ulations were performed using the Python packages QuTiP [59], 5. Litinski D. A game of surface codes: large-scale quantum
pytorch Deep Learning library for GPU [60], and pymatching [61]. computing with lattice surgery. Quantum. 2019;3:128. doi:
10.22331/q-2019-03-05-128

6. Google Quantum AI. Suppressing quantum errors by scaling

Institutional review board statement a surface code logical qubit. Nature. 2023;614(7949):676–81.
Not applicable. doi: 10.1038/s41586-022-05434-1

7. Google Quantum AI. Quantum error correction below

the surface code threshold. 2024. arXiv:2408.13687. doi:
Informed consent statement 10.48550/arXiv.2408.13687
Not applicable.
8. Jenkins A, Lis JW, Senoo A, McGrew WF, Kaufman AM.
Ytterbium nuclear-spin qubits in an optical tweezer array.
Supplementary materials Phys Rev X. 2022;12(2). doi: 10.1103/physrevx.12.021027

The supplementary materials are available at https://doi.org/10. 9. Ma S, Burgers AP, Liu G, Wilson J, Zhang B, Thompson JD.
20935/AcadQuant7467. Universal gate operations on nuclear spin qubits in an optical
tweezer array of Yb 171 atoms. Phys Rev X. 2022;12(2). doi:
10.1103/physrevx.12.021028
Additional information 10. Beugnon J, Tuchendler C, Marion H, Gaëtan A, Miroshny-
Received: 2024-09-09 chenko Y, Sortais YRP, et al. Two-dimensional transport and
transfer of a single atomic qubit in optical tweezers. Nat Phys.
Accepted: 2024-12-16 2007;3(10):696. doi: 10.1038/nphys698
Published: 2025-01-09
11. Bluvstein D, Levine H, Semeghini G, Wang TT, Ebadi S,
Academia Quantum papers should be cited as Academia Quan- Kalinowski M, et al. A quantum processor based on coher-
tum 2025, ISSN 3064-979X, https://doi.org/10.20935/Acad ent transport of entangled atom arrays. Nature. 2022;604
Quant7467. The journal’s official abbreviation is Acad. Quant. (7906):451. doi: 10.1038/s41586-022-04592-6

ACADEMIA QUANTUM 2025, 2 12 of 14

https://www.academia.edu/journals/academia-quantum/about https://doi.org/10.20935/AcadQuant7467

12. Bluvstein D, Evered SJ, Geim AA, Li SH, Zhou H, Manovitz 26. Cover TM, Thomas JA. Elements of information theory (Wi-
T, et al. Logical quantum processor based on reconfigurable ley series in telecommunications and signal processing).
atom arrays. Nature. 2024;626:58. doi: 10.1038/s41586- USA: Wiley-Interscience; 2006.
023-06927-3
27. Cong I, Levine H, Keesling A, Bluvstein D, Wang ST, Lukin
13. Nishimori H. Statistical physics of spin glasses and informa- MD. Hardware-efficient, fault-tolerant quantum computa-
tion processing: an introduction. Oxford University Press; tion with Rydberg atoms. Phys Rev X. 2022;12:021049. doi:
2001. 10.1103/PhysRevX.12.021049

14. Dennis E, Kitaev A, Landahl A, Preskill J. Topological quan- 28. Calderbank AR, Shor PW. Good quantum error-correcting
tum memory. J Math Phys. 2002;43(9):4452. doi: 10.1063/ codes exist. Phys Rev A. 1996;54(2):1098. doi: 10.1103/Phys-
1.1499754 RevA.54.1098

15. Saffman M, Walker TG, Mölmer K. Quantum information 29. Steane A. Multiple-particle interference and quantum er-
with rydberg atoms. Rev Mod Phys. 2010;82:2313–63. doi: ror correction. Proc R Soc Lond A Math Phys Eng Sci.
10.1103/RevModPhys.82.2313 1996;452(1954):2551. doi: 10.1098/rspa.1996.0136

16. Sahay K, Jin J, Claes J, Thompson JD, Puri S. High- 30. Jaksch D, Cirac J, Zoller P, Rolston SL, Côté R, Lukin
threshold codes for neutral-atom qubits with biased erasure M. Fast quantum gates for neutral atoms. Phys Rev Lett.
errors. Phys Rev X. 2023;13(4):041013. doi: 10.1103/phys- 2000;85:2208. doi: 10.1103/PhysRevLett.85.2208
revx.13.041013
31. Jandura S, Pupillo G. Time-optimal two- and three-qubit
17. Norcia MA, Young AW, Kaufman AM. Microscopic con-
gates for Rydberg atoms. Quantum. 2022;6:712. doi: 10.22
trol and detection of ultracold strontium in optical-tweezer
331/q-2022-05-13-712
arrays. Phys Rev X. 2018;8:041054. doi: 10.1103/Phys-
RevX.8.041054 32. Pagano A, Weber S, Jaschke D, Pfau T, Meinert F, Mon-
tangero S, et al. Error budgeting for a controlled-phase
18. Evered SJ, Bluvstein D, Kalinowski M, Ebadi S, Manovitz T,
gate with strontium-88 Rydberg atoms. Phys Rev Res.
Zhou H, et al. High-fidelity parallel entangling gates on a
2022;4:033019. doi: 10.1103/PhysRevResearch.4.033019
neutral-atom quantum computer. Nature. 2023;622(7982):
268. doi: 10.1038/s41586-023-06481-y 33. Levine H, Keesling A, Semeghini G, Omran A, Wang TT,
Ebadi S, et al. Parallel implementation of high-fidelity multi-
19. Xu Q, Bonilla Ataides JP, Pattison C, Raveendran N, Blu-
qubit gates with neutral atoms. Phys Rev Lett. 2019;123:170
vstein D, Wurtz J, et al. Constant-overhead fault-tolerant
503. doi: 10.1103/PhysRevLett.123.170503
quantum computation with reconfigurable atom arrays. Na-
ture. 2024;20:1084–90. doi: 10.1038/s41567-024-02479-z
34. Mohan M, de Keijzer R, Kokkelmans S. Robust control and
optimal rydberg states for neutral atom two-qubit gates.
20. Bravyi S, Cross A, Gambetta J, Maslov D, Rall P, Yoder
Phys Rev Res. 2023;5(3):033052. doi: 10.1103/PhysRevRe-
T. High-threshold and low-overhead fault-tolerant quantum
search.5.033052
memory. Nature. 2024;627:778–82. doi: 10.1038/s41586-
024-07107-7
35. Tian W, Wee WJ, Qu A, Lim BJM, Datla PR, Koh VPW, et
21. Geller MR, Zhou Z. Efficient error models for fault-tolerant al. Parallel assembly of arbitrary defect-free atom arrays with
architectures and the pauli twirling approximation. Phys Rev a multitweezer algorithm. Phys Rev Appl. 2023;19(3). doi:
A. 2013;88(1):012314. doi: 10.1103/physreva.88.012314 10.1103/physrevapplied.19.034048

22. Dankert C, Cleve R, Emerson J, Livine E. Exact and ap- 36. Covey JP, Madjarov IS, Cooper A, Endres M. 2000-times
proximate unitary 2-designs and their application to fidelity repeated imaging of strontium atoms in clock-magic tweezer
estimation. Phys Rev A. 2009;80:012304. doi: 10.1103/Phys- arrays. Phys Rev Lett. 2019;122:173201.doi: 10.1103/phys-
RevA.80.012304 revlett.122.173201

23. Jandura S, Pupillo G. Surface code stabilizer measure- 37. Tan DB, Bluvstein D, Lukin MD, Cong J. Compiling quantum
ments for Rydberg atoms. 2024. arXiv:2405.16621. doi: circuits for dynamically field-programmable neutral atoms
10.48550/arXiv.2405.16621 array processors. Quantum. 2024;8:1281. doi: 10.22331/q-
2024-03-14-1281
24. Poole C, Graham TM, Perlin MA, Otten M, Saffman M.
Architecture for fast implementation of qLDPC codes with 38. Chubb CT, Flammia ST. Statistical mechanical models for
optimized Rydberg gates. 2024. arXiv:2404.18809. doi: quantum codes with correlated noise. Ann Inst Henri
10.48550/arXiv.2404.18809 Poincaré D. 2021;8(2):269. doi: 10.4171/AIHPD/105

25. Scholl P, Shaw AL, Tsai RBS, Finkelstein R, Choi J, En- 39. Wang C, Harrington J, Preskill J. Confinement-higgs transi-
dres M. Erasure conversion in a high-fidelity Rydberg tion in a disordered gauge theory and the accuracy thresh-
quantum simulator. Nature. 2023;622(7982):273–8. doi: old for quantum memory. Ann Phys. 2003;303(1):31. doi:
https://doi.org/10.1038/s41586-023-06516-4 10.1016/S0003-4916%2802%2900019-2

ACADEMIA QUANTUM 2025, 2 13 of 14

https://www.academia.edu/journals/academia-quantum/about https://doi.org/10.20935/AcadQuant7467

40. Vodola D, Rispler M, Kim S, Müller M. Fundamental thresh- 52. Jacobsen J. High-precision percolation thresholds and
olds of realistic quantum error correction circuits from clas- potts-model critical manifolds from graph polynomials. J
sical spin models. Quantum. 2022;6:618. doi: 10.22331/q- Phys A Math Theor. 2014;47:135001. doi: 10.1088/1751-
2022-01-05-618 8113/47/13/135001

41. Edmonds J. Paths, trees, and flowers. Can J Math. 53. Kesten H. The critical probability of bond percolation on the
1965;17:449. doi: 10.4153/CJM-1965-045-4 square lattice equals 1/2. Commun Math Phys. 1980;74:41.
doi: 10.1007/BF01197577
42. Roffe J, White DR, Burton S, Campbell E. Decoding
across the quantum low-density parity-check code landscape. 54. Caracciolo S, Edwards RG, Ferreira SJ, Pelissetto A, Sokal
Phys Rev Res. 2020;2:043423. doi: 10.1103/PhysRevRe- AD. Extrapolating monte carlo simulations to infinite
search.2.043423 volume: finite-size scaling at ξ/L _ 1. Phys Rev Lett.
1995;74:2969. doi: 10.1103/PhysRevLett.74.2969
43. Honecker A, Picco M, Pujol P. Nishimori point in the 2-D +-J
random bond Ising model. Phys Rev Lett. 2001;87:047201. 55. Palassini M, Caracciolo S. Universal finite-size scaling func-
arXiv:condmat/0010143. doi: 10.1103/PhysRevLett.87.0 tions in the 3d ising spin glass. Phys Rev Lett. 1999;82:5128.
47201 doi: 10.1103/PhysRevLett.82.5128

44. Le Doussal P, Harris AB. Location of the ising spin- 56. Madjarov I. Entangling, controlling, and detecting individual
glass multicritical point on nishimori’s line. Phys Rev Lett. strontium atoms in optical tweezer arrays [PhD dissertation].
1988;61:625–8. doi: 10.1103/PhysRevLett.61.625 Pasadena (CA): California Institute of Technology; 2021.

45. Reger JD, Zippelius A. Three-dimensional random-bond 57. Wu Y, Kolkowitz S, Puri S, Thompson JD. Erasure conver-
ising model: phase diagram and critical properties. Phys Rev sion for fault-tolerant quantum computing in alkaline earth
Lett. 1986;57:3225. doi: 10.1103/PhysRevLett.57.3225 rydberg atom arrays. Nat Commun. 2022;13(1):4657. doi:
10.1038/s41467-022-32094-6
46. Gidney C. Stim: a fast stabilizer circuit simulator. Quantum.
2021;5:497. doi: 10.22331/q-2021-07-06-497 58. de Keijzer R, Visser L, Tse O, Kokkelmans S. Qubit fi-
delity under stochastic Schrödinger equations driven by
47. Kogut JB. An introduction to lattice gauge theory and spin colored noise. 2024. arXiv preprint arXiv:2401.11758. doi:
systems. Rev Mod Phys. 1979;51:659–713. doi: 10.1103/Rev 10.48550/arXiv.2401.11758
ModPhys.51.659
59. Johansson JR, Nation PD, Nori F. Qutip: an open-source
48. Elitzur S. Impossibility of spontaneously breaking local sym- python framework for the dynamics of open quantum sys-
metries. Phys Rev D. 1975;12:3978–82. doi: 10.1103/Phys- tems. Comput Phys Commun. 2012;183(8):1760–72. doi:
RevD.12.3978 10.1016/j.cpc.2012.02.021
49. Makeenko Y. Gauge fields on a lattice. In: Methods of con- 60. Paszke A, Gross S, Massa F, Lerer A, Bradbury J, Chanan G,
temporary gauge theory. Cambridge monographs on math- et al. Pytorch: an imperative style, high-performance deep
ematical physics. New York: Cambridge University Press; learning library. In Advances in neural information process-
2002. p. 99. ing systems 32. Curran Associates, Inc; 2019. p. 8024.
50. Lee JY, Ji W, Bi Z, Fisher MPA. Decoding measurement- 61. Higgott O. Pymatching: a python package for decoding
prepared quantum phases and transitions: from ising model quantum codes with minimum-weight perfect matching.
to gauge theory, and beyond. 2022. arXiv:2208.11699 [cond- 2021. arXiv:2105.13082 [quant-ph]. doi: 10.48550/arXiv.21
mat.str-el]. doi: 10.48550/arXiv.2208.11699 05.13082
51. Suzuki M. Solution and critical behavior of some “three di-
mensional” ising models with a four-spin interaction. Phys
Rev Lett. 1972;28:507. doi: 10.1103/PhysRevLett.28.507

ACADEMIA QUANTUM 2025, 2 14 of 14